Trustworthy and Scalable Federated Edge Learning for Future Integrated Positioning, Communication and Computing System: Attacks and Defenses

The emergence of Integrated Positioning, Communication, and Computing (IPC2) technology has paved the way for advanced capabilities in physical-digital spatial positioning, intelligent communication, and computing. This paper delves into an in-depth exploration of a federated learning-assisted multi-dimensionality fusion IPC2 system. Within this system, edge nodes collaboratively harness their locally distributed multi-dimensionality positioning and communication data to coordinate edge computing resources for model training. Throughout the process of fully distributed collaborative training, we focus on addressing two specific security concerns: data tampering and model tampering attacks. In pursuit of bolstering the system’s resilience against potential attacks, we introduce a novel federated-blockchain edge learning framework. This framework capitalizes on the inherent features of the blockchain, namely, its non-tampering and traceability attributes. In addition, we present a meticulously designed verification algorithm tailored for the parameters aggregation process. Specifically, an aggregation algorithm is developed to enhance the efficiency and accuracy of the training model’s fitting. To assess the effectiveness of our proposed approach, comprehensive simulations are conducted using an openly accessible wireless Artificial Intelligence (AI) dataset. The outcomes of these simulations clearly demonstrate that the proposed scheme adeptly combats data tampering attacks initiated by multiple malicious nodes and high-intensity model tampering attacks, all while maintaining minimal accuracy loss.